Understanding Visual Matching in Computer Vision

Computer vision is a discipline of computer science that plays a bigger role in your life than you think. Throughout this course, you will be introduced to the process of establishing preliminary feature matches between two or more images. This includes selecting the appropriate matching strategy for determining which correspondences are passed onto the next stage for further processing and ensuring whether the set of matching features are geometrically consistent. The procedure for devising efficient data structures and algorithms to match the features across different images is explained. You will then discover the Hough transform technique’s significance for having each edge point’s votes for all possible lines passing through it, and lines corresponding to the high accumulator or bin values are examined for potential line fits.

Next, the material explains how the Hough transform method is used to group edgels into line segments, even across gaps and occlusions in an image. You will be taught about the bag-of-features approach that computes the distribution of visual words found in the query image and compares this distribution to those found in the training images. This explanation includes the process of representing objects and images as disjointed collections of feature descriptors. The process of matching multiple images using selective match kernels is discussed. You will be taught about a match kernel that takes the best of existing techniques by combining an aggregation procedure with another particular match kernel. The various metrics used for comparing images based on their local descriptors for improving visual recognition of objects, locations and scenes will be investigated. In addition, the process of matching kernels for image pyramids is also explained.

Finally, you explore the methods of mapping unordered feature sets of images to multi-resolution histograms to find implicit correspondences based on the finest resolution histogram cell. This includes the procedure for matching two collections of features in a high-dimensional appearance space. Lastly, the fundamental principles and common pipelines of computer vision that are applied to deep neural networks are highlighted. ‘Understanding Visual Matching in Computer Vision’ is an informative course that will interest those studying computer science or artificial intelligence. Sign up today and start learning about visual matching methods in the processing of images and the methods of generating efficient and effective visual codebook techniques using additive kernels.